Parichay Kapoor
Parichay Kapoor
I am using `tensor_converter` element to convert decoded png frames to tensor format and store them into binary files. Below is the used pipeline: ``` gst-launch-1.0 multifilesrc location="frame%01d.png" ! pngdec...
Disable support for pytorch and caffe2 filters for ubuntu version < 18.04 This is because pytorch and caffe2 are not independent of the protobuf version they are compiled with. With...
Tensor elements do not handle the `GST_QUERY_LATENCY`. This query is important with live sources, and not handling this query leads to frame drops with live source/sinks. This needs all the...
TAOS-CI performs filesize check for files which are just moved (renamed as per github naming). Filesize check should not be done for renamed files.
Wanted to opinion about adding val-grind as a PR-checker. val-grind is known to have issues with false positives but glib maintains a suppression file to suppress such issues with their...
The public API requires the optimizer/dataset to be set before ml_train_model_compile. However, these restrictions can be relaxed and both can be later right until the training is being done: 1....
Adding a layer into a model created with INI requires creating two connections for the new layer - with the previous layer and with the next layer. In order to...
The current model save/load file format stores the tensors in a linear fashion. This has a couple of issues: 1. re-ordering the model layers results in failure 2. adding a...
Current inference API requires the caller to provide individual inputs to the inference call. This requires handling the inputs, outputs, and file reading on the user's end. The proposal is...
Adam variables are loaded with `ml::train::model::load()`, which allows loading the model weights. However, model weights can be loaded with a call to `ml::train::model::load()` but adam weights can only be loaded...